arm64/dma-mapping: Extend DMA ops workaround to PCI devices
authorRobin Murphy <robin.murphy@arm.com>
Tue, 19 Apr 2016 16:01:31 +0000 (17:01 +0100)
committerWill Deacon <will.deacon@arm.com>
Thu, 21 Apr 2016 16:34:34 +0000 (17:34 +0100)
PCI devices now suffer the same hiccup as platform devices, in that they
get their DMA ops configured before they have been added to their bus,
and thus before we know whether they have successfully registered with
an IOMMU or not. Until the necessary driver core changes to reorder
calls during device creation have been worked out, extend our delayed
notifier trick onto the PCI bus so as to avoid broken DMA ops once
IOMMUs get plugged into the PCI code.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
arch/arm64/mm/dma-mapping.c

index a6e757cbab7785ed411e919b95c6d13caaf21726..607e709111731184c67a58bc039f583be446597f 100644 (file)
@@ -933,6 +933,10 @@ static int __init __iommu_dma_init(void)
                ret = register_iommu_dma_ops_notifier(&platform_bus_type);
        if (!ret)
                ret = register_iommu_dma_ops_notifier(&amba_bustype);
+#ifdef CONFIG_PCI
+       if (!ret)
+               ret = register_iommu_dma_ops_notifier(&pci_bus_type);
+#endif
 
        /* handle devices queued before this arch_initcall */
        if (!ret)